Data sovereignty in investment management: the foundation for AI

Every investment manager we speak to is being told the same story: artificial intelligence will reshape the industry. They are also being told, more quietly, that the way to access it is to send their data to someone else's system. That second story is the one that should give a CIO pause. In investment management, data is not an input. It is the firm.

Data sovereignty — the principle that an investment firm retains exclusive control over its own data, its provenance and the intelligence derived from it — is becoming the most important boundary line in the technology decisions firms are making today. It is not an ideological question. It is a commercial one. The firm that does not own its data does not own its edge.

Why AI changes the calculus

For decades, investment firms have shared price data, reference data and even some trade data with vendors as a matter of course. The trade was acceptable: the vendor delivered utility, the firm retained the proprietary insight. Generative and agentic AI dissolve that boundary. A model trained on a firm's data does not return the data — it absorbs it. Every prompt sent to an external model is, in some sense, a research note shared with the provider's future capability.

This is not a theoretical concern. The intellectual property of an investment firm — its house view, its position rationale, its risk decisions — exists in the way analysts and portfolio managers express themselves in writing. If that expression is routed through an AI layer the firm does not control, the firm is contributing to a capability that its competitors will eventually rent.

What data sovereignty actually requires

Data sovereignty is not a slogan. It is an architectural choice with four practical requirements. The data must live in infrastructure the firm controls or contractually owns. The intelligence layer — the embeddings, fine-tunes and retrieval indexes that AI models depend on — must be the firm's property, not the vendor's. Access to the data and the intelligence layer must be governed by the firm's identity and entitlements stack, not the vendor's. And the firm must retain the right to extract, delete and verify deletion of its data at any time.

Firms that meet these four requirements can deploy AI aggressively, because every model improvement compounds inside the firm's own walls. Firms that do not are renting capability that they could have owned.

The application layer is becoming commoditised

A decade ago, the application layer — the OMS, the EMS, the portfolio system — was the seat of competitive advantage. That is no longer where the durable advantage lives. The application layer is converging on a small number of well-engineered options that any firm can buy. The differentiation is moving downward, into the data layer, and outward, into how a firm assembles intelligence from its data.

This is good news for investment firms. It means the firm does not need to outspend competitors on technology. It needs to be more disciplined about what data it owns, how it is structured and how AI can be brought to it without it leaving.

The vendor question to ask

When evaluating any AI-enabled vendor, the test is simple. Ask: if we terminate this contract tomorrow, what do we keep? If the answer is the raw data alone, the firm is renting. If the answer is the data, the embeddings, the retrieval indexes and the fine-tuned model weights, the firm is sovereign. Most current vendor contracts fail this test. Firms that rewrite their procurement standards now will find themselves in a materially stronger position in three years.

The strategic conclusion

AI capability will become abundant. Data sovereignty will become scarce. The investment firms that thrive in the next cycle will treat their data as the asset that it is, structure it for AI consumption inside their own boundary, and view any technology decision that erodes sovereignty as an erosion of franchise value. That is a board-level discussion, not an IT one. The firms having it now are the ones we expect to see leading their peer groups by the end of the decade.